Google DeepMind CEO Demis Hassabis: "We Will Have AGI By 2030"

Defining True Artificial General Intelligence

• 💡 Demis Hassabis sets a much higher bar for AGI than others, defining it as a system capable of all human cognitive capabilities.
• 🧠 This includes Einstein-level theory creation (inventing new physics or mathematics) and Picasso-level artistic invention (creating new art genres), not just solving existing problems.
• 🏋️‍♂️ It also encompasses physical intelligence matching elite athletes, which current robotics are far from achieving.
• 🚫 Hassabis criticizes turning AGI into a marketing term for commercial gain, emphasizing its scientific definition.

Fundamental Gaps in Current AI Systems

• ⚠️ Current AI systems, including large language models (LLMs), are "nowhere near" AGI due to fundamental limitations.
• 🔑 Key missing breakthroughs include continual learning, better memory (efficient context windows), and long-term reasoning and planning.
• 🐠 The "goldfish brain" problem means LLMs forget conversations once a session closes and do not persistently learn or change themselves.
• 💡 Hassabis believes one or two more big insightful innovations are required, beyond just scaling existing technologies.

DeepMind's Path to AGI: World Models

• 🚀 DeepMind is developing "world models", AI systems designed to understand physics, causality, and simulate reality.
• 🌍 These models, like advanced video generators, learn intuitive physics and how objects behave in the physical world.
• 🎯 World models are seen as the missing piece to enable the long-horizon planning and reasoning that humans perform effortlessly.
• 🤖 This technology is crucial for applications like robotics, allowing systems to plan trajectories and imagine scenarios in the real world.

The AGI Moment and Scientific Breakthroughs

• ✨ The true "AGI moment" will involve discovering new scientific breakthroughs that human minds haven't found.
• 🔬 Examples include finding room-temperature superconductors, new energy sources, or optimal battery designs.
• ✅ Hassabis predicts these discoveries will become possible and will happen once AGI reaches human-level knowledge and can explore uncharted territory.
• 🧬 DeepMind's AlphaFold is cited as an example of AI accelerating scientific discovery, predicting protein structures and impacting disease research.